Know issues

[dosimont]

Hi,

@ycorre: You told me about some issues and/or missing functionnalities that prevent from getting a pjdump trace containing all the original trace information, but I don't remember them.
Since I have to use this tool, I can put some efforts to improve it, and it would be better to have in mind problematic behaviors or missing features that have already been detected.

Thx!

[schnorr]

@dosimont Some work has been carried out to convert paraver trace files to two file formats: time-independent traces of SimGrid so we can replay paraver files in SimGrid; and CSV (following the pjdump specs) to visualize paraver with R+ggplot2 properly. This is being carried out in the scope of @tcbozzetti bachelor thesis. You can start here: https://github.com/tcbozzetti/trabalhoconclusao, but the prv2pjdump conversion script (strongly inspired by what Arnaud did in the past) is in here:
https://github.com/tcbozzetti/trabalhoconclusao/blob/master/getpjdump/prv2pjdump.pl
I think it is already working. Let us know if it is not the case. We are having some issues now that are described in the latest entries of @tcbozzetti LabBook (in the same repo link above), regarding asynchronous communication of MPI.

[dosimont]

Okay, I'm trying now to convert a prv trace using your script.
Is the software and hardware hierarchy represented the same way we did for our experiments with Ocelotl (threads/processes/nodes/clusters/...)?
I took a very quick look on @tcbozzetti LabBook, but it seems the problem you mention is only related to prv to SimGrid, so, it shouldn't affect me.